You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The idea would be to support CuPy just like NumPy. The missing feature is probably some kind of a "device" annotation for arrays, we would represent them in ASR and then in the backend we would use Cuda, probably using the mechanism from #1996.
sorry for the delay. Here is a very simple comparison that I did on a Tesla V100S-PCIE-32GB GPU that shows the power of cuda kernels over large arrays.
Nice. I think our (to be written) CUDA backend should be able to create an equivalent of the custom kernel, since it knows all the information at compile time to do it.
See #1964 (comment).
The idea would be to support CuPy just like NumPy. The missing feature is probably some kind of a "device" annotation for arrays, we would represent them in ASR and then in the backend we would use Cuda, probably using the mechanism from #1996.
Relevant: #1458
The text was updated successfully, but these errors were encountered: